5 AI Discovery Protocols Every Website Should Implement
AI agents don't just visit your website — they need to understand it. Before an agent can book a flight, search your inventory, or fill a form, it needs to discover what your site can do. This discovery happens through standardized protocols that act as signposts for AI systems.
Here are the five protocols that matter most in 2026, how to implement each one, and why they collectively make your site AI-native.
1. A2A Agent Cards
Google's Agent-to-Agent (A2A) protocol defines an "Agent Card" — a JSON file at /.well-known/agent.json that describes what your agent or service can do. Think of it as a business card for AI systems.
{
"name": "Acme Travel Agent",
"description": "Book flights, hotels, and car rentals",
"url": "https://acme-travel.com",
"version": "1.0.0",
"capabilities": {
"streaming": true,
"pushNotifications": false
},
"skills": [
{
"id": "flight-search",
"name": "Search Flights",
"description": "Find available flights between two cities",
"tags": ["travel", "flights"]
},
{
"id": "hotel-booking",
"name": "Book Hotels",
"description": "Search and reserve hotel rooms",
"tags": ["travel", "hotels"]
}
]
}How to add it: Create the file at /.well-known/agent.json on your web server. No special server configuration needed beyond serving static JSON. A2A Agent Cards are fetched by Google's AI systems and other A2A-compatible agents to understand your service before initiating any interaction.
2. MCP Discovery
The Model Context Protocol (MCP), originally from Anthropic, defines a server discovery endpoint at /.well-known/mcp.json. This tells AI systems where your MCP server is and what tools it offers.
{
"mcpServers": {
"acme-api": {
"url": "https://acme-travel.com/mcp",
"transport": "streamable-http",
"description": "Acme Travel booking and search API",
"tools": [
"search-flights",
"search-hotels",
"book-reservation"
]
}
}
}How to add it: If you have an MCP server (or plan to build one), this discovery file lets clients like Claude Desktop, Cursor, and other MCP-compatible tools auto-discover it. Even if you don't have an MCP server yet, listing the file shows AI agents that your site is MCP-aware.
3. agents.json
agents.json is a lightweight manifest at your site root that declares agent permissions, capabilities, and contact information. It's the emerging standard for websites to communicate their AI agent policy.
{
"agentPolicy": {
"allowAgents": true,
"allowedOperations": ["read", "search", "form-fill"],
"authentication": "none",
"rateLimit": "60/minute"
},
"contact": {
"email": "ai-support@acme-travel.com",
"documentation": "https://acme-travel.com/docs/ai-integration"
},
"tools": [
{
"name": "search",
"endpoint": "/api/search",
"method": "GET",
"parameters": {
"q": "Search query string",
"type": "flights|hotels|cars"
}
}
]
}How to add it: Create agents.json in your website's root directory. This file is increasingly checked by AI agents and browser extensions as a first signal of whether a site welcomes automated interaction. It's simple to maintain and costs nothing to implement.
4. OpenAPI Specification
If your site has an API, an OpenAPI (formerly Swagger) specification is essential. AI agents use OpenAPI specs to understand your API endpoints, parameters, authentication requirements, and response formats.
{
"openapi": "3.1.0",
"info": {
"title": "Acme Travel API",
"version": "2.0.0",
"description": "Search and book travel services"
},
"paths": {
"/api/flights/search": {
"get": {
"summary": "Search available flights",
"operationId": "searchFlights",
"parameters": [
{
"name": "origin",
"in": "query",
"required": true,
"schema": { "type": "string" },
"description": "Origin airport IATA code"
},
{
"name": "destination",
"in": "query",
"required": true,
"schema": { "type": "string" },
"description": "Destination airport IATA code"
}
]
}
}
}
}How to add it: Most API frameworks (Express, FastAPI, Rails, Django REST) can auto-generate OpenAPI specs. Serve it at /openapi.json or /api/openapi.yaml. AI coding assistants and ChatGPT plugins already consume OpenAPI specs natively. Link to it from your HTML with <link rel="service-desc" href="/openapi.json">.
5. llms.txt
llms.txt is a plain-text file at your site root that provides LLMs with a human-readable summary of your site's content and structure. It's like a README for AI — concise, structured, and designed to fit in a context window.
# Acme Travel
## About
Acme Travel is an online travel agency offering flights,
hotels, and car rentals in 50+ countries.
## Key Pages
- /flights — Flight search and booking
- /hotels — Hotel search and booking
- /deals — Current promotional offers
## API
- OpenAPI spec: /openapi.json
- Authentication: API key required for write operations
- Rate limit: 60 requests/minute
## Contact
- Support: support@acme-travel.com
- AI Integration: ai-support@acme-travel.comHow to add it: Create a plain text file called llms.txt in your site root. Keep it under 2,000 words. Focus on what your site does, its main pages, available APIs, and how AI systems should interact with it. Some sites also provide llms-full.txt with more detailed content for agents that want deeper context.
How These Protocols Work Together
Think of it in layers: llms.txt provides the overview, agents.json declares permissions, OpenAPI describes your API, mcp.json enables MCP tool discovery, and agent.json enables agent-to-agent coordination. Each serves a different audience of AI systems.
No single protocol covers everything. An AI coding assistant might check your OpenAPI spec. A browser-based agent might look for A2A Agent Cards. An LLM doing research might read your llms.txt. By implementing all five, you ensure maximum discoverability across the entire AI ecosystem.
Implementation Priority
- llms.txt — 5 minutes to create, immediately useful for all LLMs
- agents.json — 10 minutes, declares your AI policy clearly
- OpenAPI — If you have an API, this should already exist; if not, generate it
- A2A Agent Card — 15 minutes for the JSON file, positions you for Google's AI ecosystem
- MCP Discovery — Requires an MCP server, but the discovery file itself is trivial
AgentReady detects all five protocols automatically during a scan and tells you exactly which ones are missing and how to add them.
Which protocols are you missing?
Run a free scan and find out in seconds. AgentReady checks all 5 AI discovery protocols.
Scan Your Site Free